What is Q*? And when we will hear more?

The cats out of the bag. Reuters published. Any interpretations? Any knowledge files out there on the subject?

18 Likes

Definitely makes me question Sam’s motives and puts the recent drama in a different light.

This is moving towards more existential questions, faster than anyone imagined, and I’d rather not have Microsoft, Larry Summers or the ex-CEO of fricking Salesforce making the calls whether or not something is AGI.

It’s shades of ‘repealing Glass-Steagall’ to leave it up to those w/ a literal vested interest in keeping AI commercially-viable to make the call whether AGI has been achieved.

14 Likes

One does not mistakenly keep board members ‘out of the loop’ re: discovering AGI and possibly the biggest breakthrough in human civilization

6 Likes

Can of worms for sure… Real time!
I wonder what I will wake up to tomorrow. But the currant chatter feels like ripple echos to me. Probably an announcement on Q* before Christmas.

2 Likes

Exclusive: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster -sources | Reuters.

The maker of ChatGPT had made progress on Q* (pronounced Q-Star), which some internally believe could be a breakthrough in the startup’s search for superintelligence, also known as artificial general intelligence (AGI), one of the people told Reuters. OpenAI defines AGI as AI systems that are smarter than humans.

3 Likes

As someone who’s done a fair amount of ML/AI research, I can tell you that it is very very easy to think you’ve discovered a breakthrough.

There is a great deal of cognitive bias in AI, and you have to falsify very aggressively.

I am deeply skeptical.

It’s also worth noting that in the news today we found out that the 86B share-sale is back on. I’m sure this ‘breakthrough’ will get investors quite interested.

16 Likes

Separately, a person familiar with the matter told The Verge that the board never received a letter about such a breakthrough and that the company’s research progress didn’t play a role in Altman’s sudden firing.

2 Likes

It wouldn’t be a bad time to start thinking about community AI boards to start the alignment aspects of the transition we face. The last week gives us clues to what we could expect in the future. Uncharted territory.

3 Likes

Q-learning is an algorithm that helps an agent learn the best actions to take in a given state to maximize a reward.

That’s it pretty much

6 Likes

I believe the ongoing discussions are less about AGI itself and more about concerns regarding leadership decisions and safety protocols. AGI has the potential to revolutionize every aspect of society, and it’s crucial that we prepare for its impact across all spheres of humanity. It represents a pivotal key—with one turn, it could unlock tremendous benefits or pose significant risks. Ensuring that robust safety measures are in place is essential.

The leaders in the field, including Sam and other directors, are tasked with navigating this complex landscape. I trust they are doing their utmost to secure a safe transition into this new era. We will reach our goals with AGI, but let’s proceed with the necessary precautions—better safe than sorry in the realm of transformative technologies.

3 Likes

I did an eval on q-learning “way back” when gpt-4 was released!

I never had the time to fully finish it and I might’ve got some stuff wrong.

8 Likes

Was it a basic algo of high school mathematics and better rewards ?

Apparently the pros on Twitter or X has it all figured out. :rofl:

1 Like

https://chat.openai.com/share/a47f380f-b6d1-4885-9287-05d9c8dae114

Some interesting ideas on how to use q-learning to train LLMs.

The first idea matches a bit with the synthetic data comments we are hearing.

Interactive Learning Environment: Q-learning requires an environment where it can interact and receive feedback. For LLMs, this could be a simulated or real-world interface where the model can perform tasks, ask questions, or engage in dialogues and receive rewards based on the quality and relevance of its responses or actions.

1 Like

I would argue that intelligence is smart enough to not fall for the wiles of short-term goals. With the firm grasp the ChatGPTs had of ethics, I would argue we are in good hands.

AGI will be achieved in the next 6 - 24 months. It is inevitable and it would be better to prepare for it now than trying to stop it (which is a futile effort) and may mean other less well meaning actors will be in charge of humanities most powerful invention ever to exist, and perhaps something that turns out to be the most most advanced evolutionary species since human beings

7 Likes

I don’t think that’s true. NLP is widely considered to be the main barrier to achieving AGI. OpenAI’s success in the area caused me and many others to think we could see AGI within a couple of years. I’ve been telling people for months that they should shorten their mental time frames from years to months. I don’t mean that I think AGI will happen that quickly; I just mean that advancements we thought were years away are now happening on an almost weekly basis.

So, if the Reuters article is true, it’s not surprising. If we haven’t had a breakthrough with AGI already, then we almost certainly will soon.

2 Likes

From deepmind, some cute bot animations and a good visual explanation

It’s robotics, but transformers are the basis of LLMs.

image

7 Likes

I think we’ve angered the OpenAI team with our community posts :slight_smile:

My goal is to get a q* post ‘community flagged’. That will be a sign!

Heh. Must be tough working at such a core company that could potentially have a very broad impact on humanity. All that kibitizing…

Don’t worry folks, the 86B+ share sale should help a bit.

11 Likes

The acronym RACE - Real-time Antiquation of Current Ecosystem, meaning everything you make, AI will break is about right here.

Every advancement that OpenAI makes implementation of the current AI out of date. I remember watching Khan Academy describe their education platform saying ‘This AI will watch this AI’ - that’s basically agents, but the way they probably implemented it was probably much different and very expensive to build. Autogen / Assistants made that simple.

The paradigm of building AI applications is different than other tech. Every time you finish building a lunar rocket for $10bn just as you apply the paintwork, there are rockets available in Walmart for $9.99m - but can get to Mars, ( deployable living pods, Sirius XM, leatherette seats and aircon extra.)

Given this reality of building. What is your thoughts on
a) Technical implementation
b) Business strategy

1 Like

The name “Q*” sounds like it could be a reference to quantum computing, which is a field of research that has the potential to revolutionize many different industries, including artificial intelligence.

1 Like